Figure 3

Specificity performance on synthetic data. This figure shows the achieved specificities of BrierScoreMF (gray) vs linear SVM (white) on the synthetic data sets for m = 50, n ∈ {50, 150} and a varying number of classes k ∈ {2, 3, 4, 5, 10, 15, 20, 25}. For low k values, the SVM outperforms the BrierScoreMF algorithm. However, for more than 10 classes, the specificity of the linear SVM deteriorates, while BrierScoreMF achieves approximate constant prediction performance. These results were obtained by averaging 100 seeded comparisons.